Connecting Dots to Locate and Intercept…                                                 CMMC 2006      Connecting Dots to...
Connecting Dots to Locate and Intercept…                                                        CMMC 2006                 ...
Connecting Dots to Locate and Intercept…                                                 CMMC 2006the n most common or mos...
Connecting Dots to Locate and Intercept…                                                  CMMC 2006•     chemical (explosi...
Connecting Dots to Locate and Intercept…                                                  CMMC 2006technology in the every...
Connecting Dots to Locate and Intercept…                                           CMMC 2006                              ...
Connecting Dots to Locate and Intercept…                                                 CMMC 2006                        ...
Connecting Dots to Locate and Intercept…                                                 CMMC 2006Extended Kalman Filters ...
Connecting Dots to Locate and Intercept…                                                CMMC 2006become topographical feat...
Upcoming SlideShare
Loading in …5

Connecting Dots To Locate And Intercept Terrorist Operations And Operatives


Published on

  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

Connecting Dots To Locate And Intercept Terrorist Operations And Operatives

  1. 1. Connecting Dots to Locate and Intercept… CMMC 2006 Connecting Dots to Locate and Intercept Terrorist Operations and Operatives Martin Dudziak TETRAD Technologies Group, Inc. 1Our goal is to create a rather strange and non-classical kind of tool. It is not only for counteringJihad terrorism but may be applied to a variety of other problems where the patterns are skeletalor scantily clad at best, and often concealed under a cloak of mist. We want to build a map,inversely, from scattered and incomplete data points, analogous to the process of applying inversemethods to study acoustic or electromagnetic wave scattering in order to create an image wheredirect measurement is impossible for physical or other reasons. However, we are not concernedwith photons or phonons as in subsurface imaging (e.g., ultrasound), but rather with relations andsets of objects that can be treated as the virtual points making up a large network. These are thedots, and the final image is not a picture but a pattern of relations that indicates something like anoperational process – for example, a cooperating multi-cell terrorist plan to blow up airplanes inflight or trains in a subway system. However, there is a conceptual unity between the methods. Data Initial Iteration Iteration Iteration Iteration guess 3 10 20 30 (1)The abstract inverse model may be a common core for making this new kind of map, but all thebehavioral functions are different, because we are not working with Maxwell or Helmholzequations describing wave functions in a defined medium (air, glass, earth, flesh) but someinitially unclear and fuzzy rules about how the elements in the relational maps, the objects thatare people, cells of people, operational plans, travel itineraries, financial transactions, can occur.This is where we are, from the outset, still in a very speculative stage of thinking; perhaps bydescribing a few sturdy threads of investigation there emerges solid ground on which to walk.Let me begin by making some type of analogy to the use of a survey map. We want to create amap that can be bent, folded, shaken, and otherwise warped from time to time as part of theprocedure for using that map to find our way around. It is as if we were to take out an OrdnanceSurvey map and, instead of “reading” it by holding it in our hands, we were to mold it to thesurface of the earth beneath in order to find out where we are and how to get to some destination.This is a very different way of using maps which are customarily meant to be fixed in shape,retaining their internal geometry, and very accurate in scale. You don’t bend, stretch or roll upyour highway map in order to find your way from Paris to Lyon, but in this case we are aiming tohave a map that can be reshaped and morphed as we move it across a variety of possible unknownterrains in the hopes that we will get a noticeably good fit and then be able to say, “Here we areand this is some kind of valley that resembles the Loire.”1 h*ttp://,, 1
  2. 2. Connecting Dots to Locate and Intercept… CMMC 2006 (2)Consider this strange map not as a static map but as a deliberate and perpetual approximation. Infact, by assumption we know it will never be exact. We can eventually get it to fit with a type ofterrain that we do recognize because we have either seen that terrain before or it resemblessomething we have seen closely enough to be worthy of saying, “Aha! we have been herebefore.” A virtual surface, we have a topography that can serve as a guide through a terraincognito, but it is not one populated by 3D physical features. Instead there is a landscape ofrelations and patterns of relations between objects. The objects in the contextual space of terroristoperations may include cells, or individuals, or sensed physical concentrations of a chemical, apathogen, a form of radiation. They may include sequences of actions that are treated as objects –for instance, a phone call after a set of internet exchanges and then followed by a trip or themailing of a package. Ultimately in the design of the “map” we are determining what features(objects, relations) we intend to represent, and what are the scales, and the positions within acoordinate system, but it is hardly as simple as a classical map with an absolute coordinate systemand fixed locations for each object in that map. From the outset we very likely do not knowenough about the data coming to us, the mapmaker, from both human and sensor sources. In fact,if we did possess sufficient knowledge about the objects and relations, then most likely we wouldhave a clear picture and not be struggling to create an adaptive and flexible representation schemein the first place. Moreover, we have to deal with people trying to outdo and confuse themapmakers. It is a little bit like working through an iteration process and having someoneconstantly juggle the numbers on a few of the parameters by unknown degrees. Are you stillheading toward a convergence or you going askew?While in this analogy with maps and the crude representations shown above we are tempted tothink about 2D or 3D surfaces we are really faced with a higher dimensionality that depends uponthe attributes we want to consider within these patterns. Some of these objects are collections,such as groups of agents and the processes undertaken by the agents. Part of the challenge inusing (“reading”) the map is that membership in different sets can change, and we are often onlymaking estimates about the set to which a given object (actor, agent, event) should belong. Weare faced with a challenge that while sharing similarity with some of the methods employed forvolume separation and image recognition (e.g., digital Morse theory, isosurface extraction, SpiderWeb techniques), 2 there are no uniform primitives of the sort used in computerized tomographyfor instance. Or rather, we have not yet determined what are these primitives. They may bewithin reach because of the vast amount of data and conclusions about how people in smallgroups operate, the constraints in how processes like a terrorist operation can be conducted, butthe fact remains that even when we have to some satisfaction determined that we have identified2 Cox, J., Karron, D. B., Ferdous, N., “Digital Morse Theory for scalar volume data” (preprint) 2
  3. 3. Connecting Dots to Locate and Intercept… CMMC 2006the n most common or most likely configurations, we know that innovations are being pursued toconceal and encrypt and to invent alternatives that will not be detected and identified. (3)Tangible Targets Make for Better PracticeTo begin with, we create a workspace that can be a tangible, malleable subject for dialogue andbuilding an understanding by example, rather than trying to work totally in the abstract. Since weare focused upon a concrete problem of counterterrorism, we will use as a workspace a veryrealistic, in fact an exactly real scenario, that of CBRNE threats directed at mass transportation.We will attempt to work from the tangible real-world scenarios that have and may exist,backwards, inversely, toward the construction of a methodology for building inverse relationalmaps (IRMs) in the abstract which can then be used to generate massive numbers of possible fitsagainst an ever-changing dynamic and incomplete landscape of observables and inferences. Whenwe find likely fits of our flexible, stretchable maps to landscapes that are recorded as patterns thatare with high probability close to known historical scenarios, or very well-accepted andrigorously tested models of human behavior, then we will have accomplished in the domain ofterrorist modeling something that is analogous to transmitting an array of photons across asubmicron wafer surface or into a dense medium and receiving a picture that resembles a knownartifact like a smiling fetus or a hidden tunnel. (4)We have chosen to concentrate lately upon two types of threat:• biological and natural – a pandemic-potential situation such as posed by a human- transmissible H5N1 influenza virus (avian flu) 3
  4. 4. Connecting Dots to Locate and Intercept… CMMC 2006• chemical (explosive) and intentional – conventional and multi-compound liquid explosives (e.g., HMTD and TATP 3)The threat scenarios are actually closer to each other than may appear at first or second glance.Both involve from 1..m “cells” (ci), each consisting of 1..n of operative agents (pi). There areindeterminate relations between the cells (Rj, defined by (ci R ci+e)) and between agents who mayhave both intracellular (Rk, defined by (pk R pk+e)) and extracellular relations (Rk, defined by (pk(ci)R pk+e(ci))). Naturally there are differences, the most obvious being in terms of control and“cybernetic” relations that govern the actions of individual agents and cells as collective wholes.Our point here is that a pandemic virus can be disseminated by a number of cells consisting ofwild geese and ducks, passengers in a train or airplane or shoppers in a mall, and terrorist actionscan and are disseminated by cells of terrorists and their support network (both active and passive)in the population, with some common features in the dynamical structures and relations that canexist between the cells, the agents, and the kinds of processes that can occur in such networks.Common to both scenarios and a motivating force in our goal to achieve some type of practicalmodeling tools (i.e., computationally useful map generators and fitters) is the use of physicalsensors, detectors, and trackers for finding and verifying the agents and the cells. Ultimately weare looking for a certain class of relation Rq that may be intracellular or extracellular but in anycase it is an identifier, like a feature on a topographic map, that tells us with a strong probabilityvalue about an event in process. Distribution of infectious agents by birds, dogs or humans, ordistribution of explosives – or their component sets (e.g., specific acetones, acids and peroxides);either will be a feature in the map-fitting process we are trying to abstract.Once we can build up a set of relations such as an Rq that is discernable, and for which there isevidence of reproducibility, or shall we say, repeatability in both natural and manmadeoperations, then we have the beginning of some building blocks for the “internal physics” of ourinverse engine. This is the equivalent to the Maxwell/Helmholz equations that govern physicalwave scattering. This is what we can use to decide, when we have some “dots” on our evolvingmap of operations and relations detected and observed through not only sensor systems but fromhuman intelligence and inference, how the dots can move, how they can connect. Without thisbasic “virtual physics” we cannot proceed. But with it, and with sufficient data (a lot of scatteredpoints, not of light but of events including people, movement, communications, purchases,transactions), we can begin to form rough shapes in our relational space. And these rough shapesare to which we want to apply the inverse methods that have been used in optics and acoustics,such as DOT (diffusion optical tomography) and the use of level set models, so that we can see ifsome of the shapes resemble ones we have seen before, shapes that represent actions by a groupof agents or cells, for instance, in spreading a virus or in planting bombs.In our approach we have spent an increasingly large percentage of time and effort studying howto refine and perfect the sensing side of the equation, the data collection that is essential in orderto have something to put into the inverse map generation equations. This has been as necessaryas it has been in ultrasound technology development, for instance, to refine crystal transducersthat will can be used reliably in order that the signal processing can have worthy sense data.What has made the task of sensor design additionally complicated is that there are importantissues of deployment – network design and configuration from the standpoint of where and howto use sensors such as microcantilevers or SAW (surface acoustic wave) or RPAS (reversephotoacoustic spectroscopy), and also system integration from the standpoint of using any of this3 hexamethylene triperoxide diamine and (tri)acetone (tri)peroxide 4
  5. 5. Connecting Dots to Locate and Intercept… CMMC 2006technology in the everyday world of human beings - air passengers, subway riders, stadiumaudiences, all with issues of movement, privacy, and economics entering as factors.The Sensor Matrix and the Backdrop for Building Inverse Relational MapsThe recent events in the United Kingdom that could easily have led to multiple in-flight suicideterrorist attacks are indicative of the scope of threats facing the entire commercial air traffic worldas well as many other modes of public transportation and assembly. The variety of novel,accessible, and concealable forms of both liquid and solid explosives (including but not limited toHMTD, TATP and more conventional TNT, RDX and PETN (C4) compounds) offers an almostirresistible opportunity for terrorists of all types and motives to create devastation in masstransportation and situations of large public gathering. The threats are not only in the highlyvulnerable settings of airplanes in flight.While many advances have been made toward development of detection technology for use inairports, the world is now more vulnerable than ever due to three principal factors: the increasednumbers and motivational forces in terrorist networks worldwide,, due in significant part to theIraq war and prolonged hostilities therein, the increased levels of sophistication and innovationamong terrorists, and the increased availability of both supplies and communication tools, not theleast of which is through broadband and wireless internet. The logical extension of centralizedinspection systems and even measures such as the curtailment of carry-on baggage types, in thehopes of reaching a level of security that provides truly reliable security, is simply not feasiblefrom the standpoints of engineering, technology, economics or social structures. A differentapproach is necessary, and such solutions exist today and can be implemented in the near future.One pragmatic solution incorporates a particular, discrete smart card, less than the size of a creditcard, that can be used to detect multiple and necessary, critical ingredients of a large variety ofprofessional and home-made explosives, including such ingredients that are being carried orstored in advance of an attack such as on an airplane, train or bus, or inside a café or marketplace.The projected reliability of the architecture, on the basis of prior experiments and field tests, isvery high, with virtually no false negatives and very low false positives because of the emphasisupon detection and tracking of not only one specific compound but a combination of agentswithin a specific volume of space such as a suitcase or knapsack. The ergonomics andoperational cost is virtually nil and will be shown to be less disruptive in terms of passenger orcrowd movement and flow of normal business, particularly at airports and mass transit terminals.The effectiveness of the system is such that it addresses the very realistic threat of liquidexplosives and non-electronic control mechanisms that can be implemented within checkedluggage and that today poses the same risks, even greater, than those of the carry-on threatscircumvented in London and environs. The cost of each device can be low enough to allow for“use once and discard” protocols that translate into one detection device for every piece ofluggage checked or carried on any plane or train, not to speak of other applications.The figures below provide simplified views of the CEBIT (JEDI) architecture. Deployed in largenumbers, this offers not only the diversifiable, adaptive, “platform/target-independent” sensingand data collection capability needed to address today’s and tomorrow’s problems of terroristinnovation. More importantly from the standpoint of incident prevention and circumvention, thistype of sensing and data collection allows us to proceed with the type of inverse method basedprocessing that can yield maps of processes and agent networks in operation. 5
  6. 6. Connecting Dots to Locate and Intercept… CMMC 2006 (5) (6)Connecting the Dots to Make a Moldable MapIn the medical imaging world, especially that of MRI and PET for cancer imaging tasks such asare required by intensity modulated radiation therapy, there has been a vast amount of progressfrom which we believe those working on countering a different form of cancer can learn and re-use. Below we see two results of work in brain image warping used for pathogen detection. (7) 6
  7. 7. Connecting Dots to Locate and Intercept… CMMC 2006 (8)We are dealing also with surfaces and complexes that may be treated, for computational purposes,like volumetric regions. For this, DOT (diffusion optical tomography) has evolved from somebasic principles of level sets and iteration that reverses “shrinkwraps” around a true object. In theimaging case the object is something three-dimensional, onto or about which there has been wavescattering. Rayleigh Scattering Dimensions of scatterer Incident wave Back Forward are much smaller than λ scattering scattering λ Incident wave Mie Scattering Dimensions of scatterer are NOT much smaller than λ λ Incident wave (9)In our evolutionary relational maps, the objects are multi-dimensional, or they may be reducibleto three dimensions artificially, but we may still be possible to apply level set iteration concepts.In the past, the concern has been interior, boundary and exterior sets and evolving the level setover time. Miller 4 has introduced the concept of polynomial parametric level sets (PaLS) inorder to identify some characteristic function spanning the regions of interest for which all points(x, y) satisfy a relationship (e.g., p(x,y) < 0) and thus everything can be reduced to being in termsof a small set of parameters in a:4 see refs. [ - ] 7
  8. 8. Connecting Dots to Locate and Intercept… CMMC 2006Extended Kalman Filters have been thus far showing promise – ultimately this is a nonlinear leastsquares type of problem. One possibility may be something like the following, where there arethree basic elements – the data term, some form of spatial regularization, and some type oftemporal dynamics, such as a random walk, to enable a smoother and faster iteration through aseries of deformations until there is a common “core” that can be resolved to be the most stableobject.In our case the final goal of the iteration is not to obtain a final blob image that is the “trueobject” in 3D space, a clean artifact-free image of a tissue region, a mineral deposit, or a wafersurface feature. Instead the final goal for each iteration is a refinement and isolation of an objectthat represents a collection of relations Rj and Rk between cells and agents respectively, fromwhich certain relations are of type Rq, namely, critical because of resemblance to relations thathave been known from past analysis to either preview or infer certain consequences (outbreaks ofdisease, terrorist attacks).The outcome at this point is an outcome from a long train of events that start with the planningand planting of arrays and networks of sensors and observation processes, be they mobile orstatic, automated or human-operated. This brings us back to the world of CEBIT (JEDI) and thedevices we have been designing for luggage inspection, for metro stations, for airport waitinghalls, for auditoriums. However, we are actually in something of a circular process. Theplacement and organization of such a sensor network, the Deployment for Observation, is theexplicit starting point for how masses of data begin to be collected, that in turn build up thescattering points, which define the corpus of relations that we want to refine and see in clearfocus. But that planning and deployment depends upon having some idea about what is thelandscape for possible operations, terrorist or natural in origin, that can occur, that need to occur,in order for the critical events to even be possible. So we must have a prior outline to begin with.We must have a map initially, even a crude one, with which to start sketching and folding ournew map, our Inverse relational Map.And here is where we see the connection with lattices and order sets (posets) and also modelssuch as reflexive theory. All that we have been talking about concerns a way to work withsurfaces and shapes that represent the flow of information that is scattered and disjointed, eitherthrough lack of comprehensive collection, or intentional disruption and disinformation, or simplynoise and error, and we must have some guidelines, some schematic, some type of structure touse for knowing what to expect, what to look for, what to give diminished regard. The outcomesfrom using both of these techniques can produce better formulations about not only setmembership and the hierarchy of control within a given set (cell) but also rules and constraints onthe types of relations that may exist among members in a cell and between cells or amongmembers of different cells.All of these restrictive forms of information are filters that affect the range of variety we can facein terms of the relations R that will make up, or bundle themselves into, virtual objects that 8
  9. 9. Connecting Dots to Locate and Intercept… CMMC 2006become topographical features of our strange inverse mapping world. The result, hopefully, is areduction in the computation complexity that must be dealt with in the iterative diffusion-attraction “fitting” process. The outcome of an lattice-theoretic model could possibly yieldindicators of what may be the most likely set memberships for both cells and relational bundles(composites of actions, transactions, real-world processes). The outcome of a reflexive theoreticanalysis could refine these set memberships further, particularly those of relations that can orcannot likely fit together in certain spatio-temporal or cell-to-cell combinations. The next level ofanalysis is where the IRM-theoretic techniques we have been speculating about can apply, inorder to yield in the end a deformable, flexible map that can not only offer correspondences withthe real world as it has been but illuminations about what is yet to come, in time enough to dosomething about it constructively.Credits (images)(1) courtesy of Eric Miller, Northeastern University(2) (left),(right) public source unknown(4) (right) Anacostia tunnels image created by diffusion optical tomography, courtesy of EricMiller, Northeastern University(5) author(6) author(7, 8) Paul Thompson, Brain image warping and pathogen detection, K. Thomenius & B. Roysam, Introduction to Subsurface Sensing and Imaging Systems, GEGlobal Research. ECSE-4963ReferencesM. Bertero and P Boccacci. Introduction to Inverse Problems in Imaging. Institute of PhysicsPublishing (1998).D. A. Castañon , Approximate dynamic programming for sensor management, in Proc.!36th IEEEConference on Decision and Control, San Diego, CA, December (1997).D. A. Castañon, S. Streltsov, & P. Vakili, Optimality of index policies for a sequential samplingproblem, IEEE Trans. Automatic Control, 44 (1999).D. Colton, J. Coyle, & P. Monk, Recent developments in inverse acoustic scattering theory,SIAM Review, 42 (2000), pp. 369–414.A. Litman, D. Lesselier, & F. Santosa, Reconstruction of a two-dimensional binary obstacle bycontrolled evolution of a level-set, Inverse Problems, 14 (1998), pp. 685–706.E. L. Miller, M. E. Kilmer, & C. M. Rappaport, A new shape-based method for objectlocalization and characterization from scattered field data, IEEE Trans. on Geoscience andRemote Sensing: special issue on Computational Wave Issues in Remote Sensing, Imaging andTarget Identification, Propagation, and Inverse Scattering, 38 (2000), pp. 1682–1696C. R. Vogel. Computational Methods for Inverse Problems. SIAM (2002). 9